Machine CRs should have machine-parseable representations of errors

AWS does not have capacity in the [region] for [instance-type]. Please select a different instance type or region

N/A

100% of the time that AWS is at capacity (or other issues arise)

1. Create a machineset of a type of instance that AWS does not have capacity for
2. Let the machine api get the error and see it in the logs

Result is logged into the machine-api controller logs

It would be nice to see this in a condition on the machine, or some other machine-parseable format that we could hopefully either draw metrics from to better tune our alerting or to automate letting our managed-customers know that they can't spin up regions because of AWS capacity issues, or other issues.

The main thing we're looking for is a deterministic way to see these issues that is not doing something like grepping through logs.

Similarly, if possible, it would be nice to see this happen for other classes of logs. For example, if the node becomes unscheduleable for whatever reason (i.e. - autoscaling scale-down event) and the machine is being replaced, if a pod is failing to drain it would be nice to have a deterministic reason why the machine is stuck in "deleting", something we could write a script against like:

if machine is deleting:
  if machine conditions contains "type == DrainFailed"
    get machine condition where type == DrainFailed
      get Pods out of condition message -> notify customer of pods that are failing to drain